ERPs and ERP analysis

Question 1

What is an event related potential?

Answer 1

ERP is the measured brain response that is the direct result of a specific sensory, cognitive, or motor event.[1] More formally, it is any stereotyped electrophysiological response to a stimulus

Question 2

What should the trial duration in ERP be?

Answer 2

Short as possible to maximise the number of trials
Not too short to cause overlap between consecutive trials
>=1500ms if participants respond to every trial
as a rule of thumb 50 trials per participant, per condition
Add jitter of 100-300ms to avoid expectation
Choose monitor with higher refresh rate and disconnect from internet and other background programmes as it may mess up with timing

Question 3

What are ERP components and why ERP components

Answer 3

Can be physiological (circumscribed place on the scalp) or functional (circumscribed relationship to the task)
Easy to use and to analyse as the data produced is easy to interpret and link to previous findings as the activity that differs across given set of conditions is equivalent to a single component

Question 4

How to choose ERP components

Answer 4

Concentrate on one or two components at a time
Aim for large effects (ERPs with large amplitude) - easier interpretation and more robust against noise
examine well known ERPs in similar experiments to what is already studied
always compare between two conditions that differ in only one brain process
studying difference waves will eliminate common source of waveform and isolate only components that differ
Target ERPs components that are easy to isolate and have well studied difference waveforms

Question 5

What are some ERP design confounds?

Answer 5

Experimental confound - more than ONE factor differs across conditions (e.g. color and shape)
Experimental side effects - a factor varied by the experiment has side effects affecting the effect of interest
Motor confounds - one condition needs a motor response and the other one does not
Trial number confound - one condition has more trials than the other
Overlap confound - one condition is still under processing while another one is presented

Question 6

What are the rules to plot ERP?

Answer 6

Show the waveform (enough sites 6-8, to figure out the wave structure or topological scalp sites
Include pre stimulus baseline - 200ms
overlay the key waveforms
show both waveforms and their difference waveform

Question 7

How to quantify ERPs?

Answer 7

ERP amplitude - peak, mean and peak-to-peak

ERP latencies - peak, onset and factorial area

Question 8

What is peak amplitude?

Answer 8

Average around the peak or local peak amplitude.

It is done by defying a window and finding the maximum voltage in that time window = peak amplitude

Question 9

What are some problems with peak amplitude?

Answer 9

Contaminated by high frequency noise
Easily influenced by overlapping components
depends on the number of trials in the average process
Nonlinear and cannot be compared to grand - average
It is better to use mean are amplitude as the peak may be nowhere near the center of experimental effect
the peak is at different time points for different electrodes

Question 10

What are some benefits of mean amplitude?

Answer 10

Better characterisation of the components over time
It is not sensitive to high frequency noise
Can use smaller measurement window
It is linear and can be compared
The mean of Mean amplitude is equal to the mean amplitude of grand average
same applied to single trial vs averaged waveform

Question 11

How are peak and mean amplitude measured?

Answer 11

Measured with respect to baseline, therefore selection is important

Question 12

How to select a baseline?

Answer 12

Use the average of 0-200ms pre stimulus as baseline
If too short = noise; if too long = introduces signal
Baseline may differ across conditions due to overlap therefore the post stimulus amplitude measure may vary as a result of differential baselines

Question 13

What is peak-to peak amplitude?

Answer 13

Measures peak relative to an adjacent peak in the waveform
Useful for overlapping components and less noise BUT
it is difficult to interpret, cannot be sure where the change is

Question 14

What is peak Latency?

Answer 14

It is the time at which the component reaches its minimum or maximum
Has same limitations as peak amplitude
can be biased by high frequencies (needs filtering)
Use a peal measure ( not considered a peak unless 3-5 sample points on each site have smaller values )

Question 15

What is onset latency?

Answer 15

It is the latency (time) of the initiation/onset of a process
Tricky to estimate
The onset of a difference between conditions is the point at which the difference is just greater than zero
Regression line is fitted to the baseline and another or rising/falling ERP response.
The onset is the intercourse between the two lines

Question 16

What is exploratory ERP analysis?

Answer 16

at each sample point - t-test between ERPs
retain only time points successive for a certain period
Useful to find ROIs for subsequent analysis

Question 17

How to exclude participants of the ERP analysis?

Answer 17

Established priori based on a criteria:

• did they performed the task (behaviour exclusion - e.g. exclude if performance is under a specified standard)
• do the waveforms manifest the component of interest

Question 18

How to report participants exclusion?

Answer 18

Provide the details of exclusion and the criteria (no cherry picking)
Present waveforms for excluded participants to show the validity
Or present data with and without participants

Question 19

Why is quantifying an ERP waveform used for?

Answer 19

It is exploratory
Appropriate for research driven by effect-unspecified hypothesis (e.g. assuming a difference but not aware how the processes differ)

Question 20

What are the steps of the basic approach of quantifying ERP waveform?

Answer 20

Divide the wave into series of time windows
calculate mean amplitude measures within each window
statistical test is done for each window
differences in the waveforms are quantifies in relations to window showing p-values
problems of multiple comparison (type 1 error)
limited theoretical value

Question 21

What are the two spatial derivatives?

Answer 21

Gradient

Current Source density (CSD)

Question 22

What is Gradient?

Answer 22

First special derivative - the degree of change the voltage measured at electrode X relative to its neighbours
Vectors that vary in strength and direction
Bipolar recording

Question 23

What is current Source Density (CSD)?

Answer 23

Second special derivative - the degree of change of the degree of change of the voltage in X relative to its neighbours
Conceptually similar to special high pass filter
Enhances and separates focal activity and removes low special frequency assumed to originate from deeper sources

Question 24

What are mapping descriptors?

Answer 24

No data reduction, increases the amount of data through interpolation
Cannot infer if changes are global or local by looking at electrodes in isolation
useful to have map descriptors which are sensitive to change in the spatial distribution but insensitive to global strength and vice versa

Question 25

What is global field power (GFP)?

Answer 25

Related to root mean square
Reference free
plotted as a function of time
Nonlinear transformation - GFP of the mean ERP is not equal to the mean of the GFP across participants
Tells us on average how strong a potential has been recorded across the electrodes montage
Does not give information about the potential distribution across the montage (where small or large potentials are measured)

Question 26

What is global map dissimilarity (GMD)?

Answer 26

Equivalent to special correlation coefficient between two maps
measures topographies differences between two maps
0 for identical maps and 2 for maps of opposite polarities
Inversely correlated with GFP
It defines if different sources are involved in generating the electrical activity at the two scalp maps
Permutation tests to compare GMDs

Question 27

What is ERP source analysis?

Answer 27

It is used for localisation which is important and main component of neuroscience
Not suitable as a principal analysis but may ne useful as additional/exploratory analysis

Question 28

What is the forward problem?

Answer 28

Given a source what are the scalp waveforms
we have the dipoles in specific regions and want to find out the topological map
It has an unique solution
Can be solved analytically without EEG data
Accuracy depends on how is the head modeled anatomically accurate or sphere; it assumes standard electrode montage
Localisation can be done with two dipoles as long as they are far apart and superficial
single superficial dipole (focused scalp distribution, when shifted the distribution changes a lot, easy to localise)
deeper dipole (broader scalp distribution, changes in dipole have small impact on the scalp distribution, harder to localise and to distinguish from superficial activity)

Question 29

What is the inverse problem?

Answer 29

Given the scalp distribution what is the source
does not have a unique solution but infinite number of solutions for any observed voltage distribution
If there is noise in the data, the correct solution may be substantially different from the best fit solution

Question 30

What are the characteristics of the Equivalent current Dipole approach?

Answer 30

Small number of equivalent dipoles <10
fixed locations and orientations, with varying magnitude
compares the forward solution and observed distribution over a range of time
finds the locations and orientations that together provide the best fit over time
Each dipole has 5 parameters plus magnitude (need to be careful with choice
The 5 parameters consist of
3 x location and 2 x orientation

Question 31

What are some strategies to use in Equivalent Current Dipole approach (ECD)?

Answer 31

Can use PCA to determine the number of underplaying special distributions of activity
Start with a few sensory dipoles to fit early ERPs and then add more later on
Use preexisting strategies

Question 32

What are some limitations of ECD?

Answer 32

Dependent on the operator - can be biased by expectations as has 5 free parameters to choose
Difficult to asses the accuracy of the solution - 5 free parameters per dipole and if one is wrong other variables slightly change to maintain low variance; if there is noise the wrong solution may have lower residual variance than the correct
Cannot make priori predictions about the number of dipoles
Most activity is distributed over large body of cortex

Question 33

What is distributed source imaging?

Answer 33

Divides the brain in small number of voxels and finds pattern of activation values that produces the observed data
3 dipoles in each voxel
Needs a large number of free parameters to be estimated

Question 34

What are the approaches to distributed source imaging and what do they mean?

Answer 34

Minimum Norm - provides a unique solution (may not be the best), but is biased towards deep sources as they must be strong to have an impact on the scalp and that is what minimum norm is weighted against
Deep-weighted minimum norm - provides slightly better solution than minimum norm but still may not be the best

Question 35

Low resolution electromagnetic tomography (LORETA)

Answer 35

Chooses the maximally smooth solution

Question 36

What are some advantages of LORETA?

Answer 36

fast to compute
can be applied to fewer data points
can produce fMRI like activation maps
relatively few parameters to choose

Question 37

What are some disadvantages of LORETA?

Answer 37

Weights for each electrodes are not fine tuned to the statistical properties of the data
multiple comparison problem

Question 38

Source location

Answer 38

Better to be used by experienced EEG researchers
Take general approach to hypothesis testing
Play around in the context of discovery - localisation may lead to a hypothesis not a solution

Question 39

What is adaptive distribution source imaging (ADSI)?

Answer 39

Weights are computed against the physical electrode position and the data recorded from the electrodes
Can be computed only over time period as uses a covariance matrix

Question 40

What are some advantages of adaptive distribution source imaging?

Answer 40

weight changes over time and therefore provides increased sensitivity to detect subtle features

Question 41

What are some disadvantages of adaptive distribution source imaging?

Answer 41

Many steps and many free parametres to be choices and data is sensitive to these choices

Question 42

What are the characteristics of oscillatory signals

Answer 42

frequency (f) - number of cycles per unit of time
amplitude (A) - magnitude of displacement
phase (O) - initial displacement at time = 0

Question 43

How are oscillations represented/ in what domains ?

Answer 43

Time domain - how does the signal changes over time?
Frequency domain - how much of the signal lies within each frequency band over a range of frequencies
Power spectrum - defined periodogram from the Fourier coefficients (cannot characterise non linear patterns, equivalent to a histogram in the frequency domain)